We examine peer effects in early education by estimating value-added models with school fixed effects that control extensively for individual, family, peer, and teacher characteristics to account for the endogeneity of peer group formation. We find statistically significant and robust spillover effects from preschool on math and reading outcomes, but statistically insignificant effects on various behavioral and social outcomes. We also find that peer externalizing problems, which most likely capture classroom disturbance, hinder cognitive outcomes. Our estimates imply that ignoring spillover effects significantly understates the social returns to preschool.